Abstract:

Massive Multiple-Input Multiple-Output (MIMO) is a key enabler for 5G and future 6G networks, providing very large spectral and energy-efficiency gains by exploiting large antenna arrays at the base station. However, accurate and real-time performance optimization remains challenging due to the high dimensionality of the channel state, nonlinear interactions between system parameters, and the prohibitive computational cost of exhaustive Monte Carlo (MC) Bit Error Rate (BER) simulations. Classical analytical BER approximations are either too simplistic to capture realistic propagation conditions or too computationally expensive to be used in tight optimization loops. To address this challenge, this paper proposes an ensemble-based surrogate modeling framework for BER prediction in Massive MIMO uplink systems, combined with a surrogate-driven search for an optimal steering (or effective) angle. 



A gradient-boosted regression-tree ensemble (LSBoost-type model) is trained on a large-scale, high-fidelity MC dataset containing $27\,000$ Massive MIMO configurations and is then used as a fast BER surrogate inside an angle-optimization routine. The surrogate yields near-MC accuracy (in a BER-dB sense) while reducing evaluation time by orders of magnitude. The key novelty lies in the design of this BER surrogate as a computationally efficient replacement for direct MC BER simulations and in its integration into a surrogate-driven angle-optimization loop for Massive MIMO. Numerical results obtained with a 5G/6G-inspired channel model based on 3GPP TR~38.901 demonstrate high prediction accuracy (100\% of points within $\pm 1$~dB of the MC BER), significant speedups, and near-optimal angle selection across a wide range of system settings.